NFT @ 11282025: Fitting Google Cluster Traces for Data-Driven Analysis of Queueing Systems скачать в хорошем качестве

NFT @ 11282025: Fitting Google Cluster Traces for Data-Driven Analysis of Queueing Systems

#researchatnecstlab - #NECSTFridayTalk On Friday, November 28, 2025, we had a new talk for the series #NECSTFridayTalk. During this talk, we had, as speaker, Thu Le-Anh, PhD at University of Tsukuba. In the following, you can find the details about the talk: 📌 Title: Fitting Google Cluster Traces for Data-Driven Analysis of Queueing Systems with Setup Policies 📌 Abstract: The rapid expansion of cloud computing driven by artificial intelligence and machine learning (AI/ML) applications has led to increasing workloads and energy demands in data centers. Designing efficient and sustainable systems requires accurate modeling and performance analysis, yet this remains challenging due to highly variable workloads and complex server interactions. Queueing models offer an effective framework for analyzing data center operations and evaluating key performance metrics, including response time, utilization, and energy consumption. In particular, queueing models with setup policies, where idle servers are turned off and reactivated upon the arrival of new jobs, are widely used to reduce power consumption without significantly degrading performance. These policies are particularly relevant for modern data centers that employ dynamic power-saving strategies during periods of low load. Our study investigates queueing models with setup policies by fitting interarrival times obtained from Google cluster traces. We address a common limitation of existing queueing models, which typically assume Poisson arrivals (i.e., exponentially distributed interarrival times) for analytical tractability. By capturing realistic workload dynamics and integrating them into analytical models, the proposed approach enables a more accurate evaluation of the setup policies, providing practical insights into balancing energy efficiency in large-scale data centers. #NECSTLab #Computerscience